We present a new image indexing and retrieval method for imagedatabase applications. We introduce a novel image content descriptionfeature termed scene structural matrix (SSM). The SSM captures theoverall structural information of the scene by indexing the geometricfeatures of the image. We also introduce a constraint adaptive imagesegmentation method termed middle cut (MC) to partition the image andderive the multi-scale geometric structures of the image. Experimentalresults show that SSM is particularly effective in retrieving imageswith strong structural features, such as landscape photographs. It isalso shown that SSM is robust against spatial and spectral distortionsthus making it superior to the traditional colour histogram, colourcorrelogram and MPEG7-color structure histogram in certain applications
展开▼